# Save the final merged dataframe
result <- read.csv("results/benchmarking/deseq/comp_adj/parallel_models/merged.csv", row.names = 1)
pal <- c("#969696",
"#fcc5c0", "#fa9fb5", "#f768a1", "#c51b8a",
"#addd8e", "#78c679", "#31a354", "#006837",
"#fd8d3c","#e6550d"
)
# Modify legend text
models.legend <- list(
unadjusted = ~ "Group",
raw_cm = ~ "Group + CM prop",
raw_cm_fb = ~ "Group + CM prop + 1x prop of minor cells",
raw_cm_fb_im = ~ "Group + CM prop + 2x prop of minor cells",
raw_4 = ~ "Group + CM prop + 3x prop of minor cells",
clr_cm = ~ "Group + CLR of CMs",
clr_cm_fb = ~ "Group + CLR of CMs + 1x CLR of minor cells",
clr_cm_fb_im = ~ "Group + CLR of CMs + 2x CLR of minor cells",
clr_4 = ~ "Group + CLR of CMs + 3x CLR of minor cells",
pc1 = ~ "Group + PC1",
pc2 = ~ "Group + PC1 + PC2"
)
p.fp <- result |>
ggplot(aes(x = sub_sublist_name, y = false_pos +0.01, color = sublist_name)) +
geom_point(alpha = 0.4) + # smaller and more transparent points
geom_smooth(se = F, method = "loess", size = 1.5, alpha = 0.2, span = 0.3) +
#scale_y_continuous(trans = "log2") +
scale_color_manual(values = pal,
name = "Model design",
labels = models.legend,
breaks = names(models.legend)) +
theme_minimal() +
theme(
legend.position = "none",
legend.title = element_blank(),
text = element_text(size = 12),
panel.grid.major = element_blank(), # remove major gridlines
panel.grid.minor = element_blank() # remove minor gridlines
) +
#geom_magnify(from = c(-0.07, 0.07,-3,30), to = c(-0.08,0.08, 380,600),
# shape = "rect", shadow = F) +
labs(y= "False Positives in 500 genes", x = "Simulated CM proportion difference") +
ggtitle("False positive rate in simulated dataset\n(25mil counts, mean major prop is 0.5, 5 replicates)")
p.tp <- result |>
ggplot(aes(x = sub_sublist_name, y = true_pos + 0.01, color = sublist_name, group = sublist_name)) +
geom_point(alpha = 0.2) + # smaller and more transparent points
geom_smooth(se = F, method = "loess", size = 1.5, alpha = 0.2, span = 0.3) +
#scale_y_continuous(trans = "log10") +
scale_color_manual(values = pal,
name = "Model design",
labels = models.legend,
breaks = names(models.legend)) +
theme_minimal() +
theme(
legend.position = "none",
legend.title = element_blank(),
text = element_text(size = 12),
panel.grid.major = element_blank(), # remove major gridlines
panel.grid.minor = element_blank() # remove minor gridlines
) +
#geom_magnify(from = c(-0.07, 0.07,-3,30), to = c(-0.08,0.08, 380,600),
# shape = "rect", shadow = F) +
labs(y= "True Positives out of 100 DE genes", x = "Simulated CM proportion difference") +
ggtitle("True positive rate (2x fold change)")
# Want to take the ratio of the false postive rate over the true positive rate
result <- result %>%
mutate(distance_bin = floor(abs(sub_sublist_name) / 0.05) * 0.05)
rates.df <- result %>%
mutate(
false_neg = total_pos - true_pos,
true_neg = 500 - (true_pos + false_pos + false_neg),
Precision = true_pos / (true_pos + false_pos),
Recall = true_pos / (true_pos + false_neg),
"F1 score" = 2 * (Precision * Recall) / (Precision + Recall),
Specificity = true_neg / (true_neg + false_pos)
)
# Melt the data for plotting
rates.df <- rates.df %>%
pivot_longer(cols = c(Precision, Recall, "F1 score", Specificity),
names_to = "Metric",
values_to = "Value") %>%
group_by(Metric, sublist_name) %>%
mutate(median_value = median(Value, na.rm = TRUE) ) %>% #
arrange(Metric, median_value)
p.stats <- ggplot(rates.df, aes(x = sublist_name, y = Value)) +
geom_boxplot(aes(fill = sublist_name)) +
scale_fill_manual(values = pal,
name = "Model design",
labels = models.legend,
breaks = names(models.legend)) +
facet_wrap(~ Metric, scales = "free") +
theme_minimal() +
theme(
legend.position = "right",
strip.background = element_blank(),
strip.text.x = element_text(size = 12),
axis.text.x = element_text(size = 7, angle = 30)
) +
labs(
title = "Performance Metrics by Model Type",
y = "Metric Value",
x = "Model and Sublist"
) +
scale_x_discrete(limits = unique(rates.df$sublist_name))
# Filter the data for F1 score
result_f1 <- rates.df %>%
filter(Metric == "F1 score")
# Create the scatterplot
p.f1 <- ggplot(result_f1, aes(x = abs(sub_sublist_name), y = mean_value)) +
geom_point(aes(color = sublist_name), size = 4, alpha = 0.2) +
geom_smooth(aes(color = sublist_name), se = F, method = "loess", size = 1.5, alpha = 0.3, span = 0.3) +
scale_color_manual(values = pal,
name = "Model design",
labels = models.legend,
breaks = names(models.legend)) +
theme_minimal() +
labs(
title = "F1 Score by Absolute Percent Difference",
x = "Absolute Percent Difference",
y = "F1 Score"
)
design <- c("AC
BC")
wrap_plots(A = p.fp, B = p.tp, C = p.stats, design = design)
## `geom_smooth()` using formula = 'y ~ x'
## `geom_smooth()` using formula = 'y ~ x'
pal <- c("#969696",
"#fcc5c0", "#fa9fb5", "#f768a1", "#c51b8a",
"#c6dbef", "#9ecae1", "#6baed6", "#3182bd",
"#fd8d3c","#e6550d")
# Modify legend text
models.legend <- list(
unadjusted = ~ "Unadjusted",
raw_cm = ~ "CM prop",
raw_cm_fb = ~ "CM prop + 1x",
raw_cm_fb_im = ~ "CM prop + 2x",
raw_4 = ~ "CM prop + 3x",
clr_cm = ~ "CLR of CMs",
clr_cm_fb = ~ "CLR of CMs + 1x",
clr_cm_fb_im = ~ "CLR of CMs + 2x",
clr_4 = ~ "CLR of CMs + 3x",
pc1 = ~ "PC1",
pc2 = ~ "PC1 + PC2"
)
p.fp <- result |>
subset(ModelType != "Isometric log ratios") |>
ggplot(aes(x = sub_sublist_name, y = false_pos +0.01, color = sublist_name)) +
geom_point(alpha = 0.1) + # smaller and more transparent points
geom_smooth(se = F, method = "loess", size = 1.5, alpha = 0.2, span = 0.8) +
#scale_y_continuous(trans = "log2") +
scale_color_manual(values = pal,
name = "Model design",
labels = models.legend,
breaks = names(models.legend)) +
theme_minimal() +
theme(
legend.position = "none",
legend.title = element_blank(),
text = element_text(size = 24),
panel.grid.major = element_blank(), # remove major gridlines
panel.grid.minor = element_blank() # remove minor gridlines
) +
#geom_magnify(from = c(-0.07, 0.07,-3,30), to = c(-0.08,0.08, 380,600),
# shape = "rect", shadow = F) +
labs(y= "False Positives in 500 genes", x = "Simulated CM proportion difference") +
ggtitle("False Positive Rate")
p.tp <- result |>
subset(ModelType != "Isometric log ratios") |>
ggplot(aes(x = sub_sublist_name, y = true_pos + 0.01, color = sublist_name, group = sublist_name)) +
geom_point(alpha = 0.1) + # smaller and more transparent points
geom_smooth(se = F, method = "loess", size = 1.5, alpha = 0.2, span = 0.7) +
#scale_y_continuous(trans = "log10") +
scale_color_manual(values = pal,
name = "Model design",
labels = models.legend,
breaks = names(models.legend)) +
theme_minimal() +
theme(
legend.position = "none",
legend.title = element_blank(),
text = element_text(size = 24),
panel.grid.major = element_blank(), # remove major gridlines
panel.grid.minor = element_blank(),
legend.key.width = unit(3,"cm"),
legend.key.height =unit(1,"cm"), # remove minor gridlines
) +
#geom_magnify(from = c(-0.07, 0.07,-3,30), to = c(-0.08,0.08, 380,600),
# shape = "rect", shadow = F) +
labs(y= "True Positives out of 100 DE genes", x = "Simulated CM proportion difference") +
ggtitle("True Positive Rate")
# Modify the ggplot for False Positives (p.fp) to include dodging by 'distance_bin'
p.fp.bin <- result |>
subset(ModelType != "Isometric log ratios") |>
ggplot(aes(x = sub_sublist_name, y = false_pos + 0.01, color = sublist_name)) +
geom_point(aes(dodge = distance_bin), position = position_dodge(0.2), alpha = 0.1) + # dodging
geom_smooth(se = F, method = "loess", size = 1.5, alpha = 0.2, span = 0.8) +
scale_color_manual(values = pal,
name = "Model design",
labels = models.legend,
breaks = names(models.legend)) +
theme_minimal() +
theme(
legend.position = "none",
text = element_text(size = 24),
panel.grid.major = element_blank(),
panel.grid.minor = element_blank()
) +
labs(y= "False Positives in 500 genes", x = "Simulated CM proportion difference") +
ggtitle("False Positive Rate")
p.stats <- rates.df |>
subset(ModelType != "Isometric log ratios" &
Metric == "F1 score") |>
ggplot(aes(x = factor(sublist_name, level = names(models.legend)),
y = Value)) +
geom_boxplot(aes(fill = sublist_name)) +
geom_jitter(width = 0.1, height = 0, alpha = 0.1) +
scale_fill_manual(values = pal,
name = "Model design",
labels = models.legend,
breaks = names(models.legend)) +
#facet_wrap(~ distance_bin) +
theme_minimal() +
theme(
legend.position = "bottom",
axis.text.x = element_blank(),
text = element_text(size = 24)
) +
labs(
title = "Model Accuracy",
y = "F1 Score (higher is more accurate)",
x = "Model Type"
)
design <- c("AB
CC")
wrap_plots(A = p.fp, B = p.tp, C = p.stats, design = design)
## `geom_smooth()` using formula = 'y ~ x'
## `geom_smooth()` using formula = 'y ~ x'
rates.df |>
subset(ModelType != "Isometric log ratios") |>
ggplot(aes(x = factor(sublist_name, level = names(models.legend)),
y = Value)) +
geom_boxplot(aes(fill = sublist_name)) +
geom_jitter(width = 0.1, height = 0, alpha = 0.1) +
scale_fill_manual(values = pal,
name = "Model design",
labels = models.legend,
breaks = names(models.legend)) +
facet_wrap(~ Metric, scales = "free") +
theme_minimal() +
theme(
legend.position = "bottom",
axis.text.x = element_blank(),
text = element_text(size = 24)
) +
labs(
title = "Model Accuracy",
y = "F1 Score (higher is more accurate)",
x = "Model Type"
)
